Overview
Catchpoint's advanced data analytics engine can help you identify potentially significant changes in metrics by automatically detecting trend shifts. Whereas conventional alerts or events are triggered based on simple metric thresholds, trend shifts are detected using an algorithm that compares data against historical baselines to identify even subtle shifts in performance.
How Trend Shifts are Detected
Catchpoint uses a proprietary algorithm to identify trend shifts for various metrics across multiple products.
The trend shift detection algorithm is applied to historical data using a rolling timeframe. For each point in time, the algorithm compares metric data from a prior set of datapoints or records with data from the same number of records following it. We perform two different comparisons: one using longer series of data points to detect slower shifts, and one using shorter series to detect faster shifts. The algorithm gives weight to both the average of the metric in each set and the standard deviation (a measure of how much variance there is within the data). A trend shift is indicated when we calculate a shift of 20% or more for fast shifts and 10% or more for slow shifts.
Our goal in tuning this detection algorithm was to make it sensitive enough to identify significant trend shifts in most cases, without being so sensitive as to return excessive false positives. Of course, there is a level of subjectivity in defining what is a "significant" shift, but trend shift detection should provide most users valuable guidance as they analyze their Catchpoint data.
Where Trend Shifts are Used
Catchpoint's trend shift detection is automatically applied to your data in Smartboard, where any detected trend shifts will be listed in the Key Events section. In Smartboard, trend shifts are identified using the Ping RTT metric for Ping and Traceroute tests, and the Test Time metric for all other test types. Note that in Smartboard, trend shifts are detected per node (data from multiple nodes is not aggregated for trend analysis.)
Trend shifts are also used to detect outages in Internet Sonar and Stack Map.
Finally, you can configure your own Advanced Alerts to be triggered based on trends shifts, rather than fixed metric values. Learn more about Trend Shift Alerts